4.3 Factors Limiting Flux and Wavelength Accuracy

The COS calibration accuracies are given in Table 1.3 of the COS Instrument Handbook. In this section, several factors limiting those accuracies are discussed.

4.3.1 Flux Accuracy

The accuracy of the absolute flux calibration of COS spectroscopic data is limited by several factors including:

• The presence of fixed pattern noise in the FUV detectors. Although the grid wire shadows are corrected, several other artifacts remain, some of them with amplitudes up to 10%. Because the flux calibration is intended to be a smooth function, it interpolates through such small scale features, which can result in localized errors of ±5% (see COS ISR 2010-02 and COS ISR 2011-03).
• For the G140L settings, the initial wavelength calibrations were not as accurate as they are now. This, coupled with the fact that the instrumental response drops rapidly below about 1200 Angstroms, and that individual FUVB observations may not be properly aligned (see below), results in rather large flux calibration uncertainties (5–10%) below 1200 Angstroms for G140L data.
• The time-dependent sensitivity (TDS) correction to the FUV flux calibration may not be exact because it is determined empirically from available calibration observations and is determined after-the-fact as new calibration observations become available. The FUV sensitivity varies with time, and the rate depends on the grating, segment, and wavelength region considered. Based on our long baseline for COS, the degradation is likely dominated at early times by an outgassing product and at later times by atomic oxygen in the residual Earth atmosphere at HST altitude. Regular monitoring of the time-dependent sensitivity captures changes in the sensitivity due to varying atmospheric conditions stemming from variations in the solar cycle.  The decline at early times is characterized in COS ISR 2011-02; the complex behavior at early times may add an additional small error to the calibration of data early in the mission. Information about the current time-dependent sensitivity monitoring and modeling is described in COS ISR 2022-08.
• Due to on-board Doppler corrections, a given pixel in ACCUM data will contain data from nearby pixels, which will cause a slight smearing of the fixed pattern noise.
• Because no PHA filtering is done onboard, FUV ACCUMs include events for all PHA values. This has two minor effects. First, background counts are included. However, since objects observed in ACCUM mode are bright, this should not be a practical issue. Second, because the absolute flux calibrations are derived from PHA filtered TIME-TAG data, this can result in small, systematic effects in the flux calibration, but these should be less than a few percent. In addition, ACCUM data do not include walk corrections.
• Both of the COS low resolution gratings are affected by order overlaps. For the NUV G230L, wavelengths longer than about 3200 Å (which primarily affects the NUVB stripe of CENWAVE 3360), second order light from wavelengths longer than 1600 Å can contaminate the result. For the second order G230L spectra (stripe NUVC), first order light from wavelengths at twice the observed wavelength can affect the spectra (see COS ISR 2010-01). For the FUV G140L, spectra longward of about 2150 Å can be contaminated by an overlapping second order spectrum (see COS ISR 2010-02). The exact extent of the contamination depends on the SED of the object being observed.
• For low signal-to-noise observations (particularly for faint sources), the signal-to-noise is best determined by measuring the dispersion around the continuum. At high signal-to-noise, this will match what can be determined from the ratio of the flux and error arrays in the x1dsum files.

4.3.2 Wavelength and Spectral Resolution Accuracies

There are several issues that may affect the COS wavelength calibration and spectral resolution, and these are explained in detail in COS ISRs 2010-05, 2010-06, 2009-01, and 2010-09. Some of these issues are outlined here.

• Because the COS optics corrects the HST spherical aberration after light passes through its large aperture, it accepts all of the uncorrected light from the HST telescope beam. Consequently, its image quality is subject to mid-range polishing errors which create broad wings on the PSF (see COS ISR 2009-01). Other spectrographs such as STIS can eliminate the effects of these wings by inserting a small aperture into the beam. Because COS cannot do that, its spectral purity is affected by the wings.
• Small, localized deviations from the dispersion relations determined by a low order polynomial have been reported for FUV XDL data. These deviations most probably result from localized inaccuracies in the geometric correction.
• For the FUVB segment of the G140L CENWAVE = 1280 setting, the wavecal lamp does not have detectable lines. As a result, the wavelength calibration from the FUVA side is applied to FUVB. However, for some observations, the FUVA is turned off, to avoid an over-bright condition. In these cases, a default wavelength calibration is applied. Note that the wavecal not only affects the wavelength calibration itself, but also the determination of where the PSA or BOA spectrum is expected to be. These same comments apply to FUVB observations obtained with the G130M CENWAVE = 1055 and 1096 settings.  For more information, see section 5.7.5 of the COS Instrument Handbook.
• OSM motions, or drifts, can cause the spectrum to shift in the dispersion direction by as much as 2–3 pixels (~1 resolution element for NUV, approximately one-half resolution element for FUV) in the first 20 minutes after an OSM is moved. TAGFLASH wavecals correct for these motions to accuracies ≤0.5 pixel. However, it is only possible to correct ACCUM data for the mean OSM motion that occurred during the exposure and, in rare circumstances, this may result in a slight degradation in the spectral resolution of ACCUM  data.
• The accuracy to which the source is centered in the science aperture along the dispersion direction can result in small displacements in the absolute wavelength scale corresponding to the plate-scales of 0.22 arcsec per FUV pixel and 0.25 arcsec per NUV pixel. Measurements for ACQ/IMAGE centering accuracies are of the order of 0.05 arcsec, and accuracies of other types of COS acquisition can be of the order of 0.1 arcsec or more. One can calculate the resulting wavelength accuracy using the plate-scale and dispersion given in Table 1.4 and Table 1.1 respectively.
• As discussed in the COS Instrument Handbook, the BOA degrades the target image, resulting in a reduction of the spectral resolution by a factor of three or more.
• Exposures longer than 960s taken at LP6 with SPLIT wavecals have a shift applied to the wavecal calibration to account for the fact that wavecals cannot be taken at the same time as the science observations. This shift makes the wavelength calibration more accurate than a straight line time interpolation between the SPLIT wavecals, and is estimated to result in an increase of < 0.5 pixel in the uncertainty of the wavelength calibration.